菜单
Test-Driven Development debug workflow combined with multi-agent collab-fix. Use when the user asks to "TDD debug", "reproduce and fix with a test", or wants a bug fixed with a failing test written first and verified after the fix.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Revise an academic paper sentence by sentence with the author, stripping AI writing tells (stop-slop) while preserving the author's own voice. Use this whenever the user wants to polish, tighten, edit, proofread, or "de-slop" a paper, manuscript, draft, abstract, or thesis chapter — especially when they want careful interactive sentence-level revision rather than a one-shot rewrite. Trigger on requests like "help me revise my paper", "clean up my draft", "make my writing sound less AI", "go through my abstract line by line", or when they hand over a .tex/.md/.txt paper to improve.
Wield Google's Gemini CLI as a powerful auxiliary tool for code generation, review, analysis, and web research. Use when tasks benefit from a second AI perspective, current web information via Google Search, codebase architecture analysis, or parallel code generation. Also use when user explicitly requests Gemini operations.
Scan this machine for the Miasma / Phantom Gyp ("TeamPCP") npm supply-chain compromise that plants backdoors in Claude Code (~/.claude/settings.json) and VS Code (.vscode/tasks.json). Use when the user asks to check if their PC/machine is infected, compromised, or affected by the npm/Claude Code backdoor attack, credential-stealing worm, or Shai-Hulud campaign.
Remove AI writing patterns from prose. Use when drafting, editing, or reviewing text to eliminate predictable AI tells.
Use when the user asks to run Codex CLI (codex exec, codex resume) or references OpenAI Codex for code analysis, code review, or automated editing
Collaborative multi-agent fix workflow using codex, gemini-cli, and an independent subagent. Use when the user asks to "collab fix", "multi-agent fix", or wants several AI tools to jointly analyze and fix a bug or task with cross-review.
| name | TDD-debug |
| description | Test-Driven Development debug workflow combined with multi-agent collab-fix. Use when the user asks to "TDD debug", "reproduce and fix with a test", or wants a bug fixed with a failing test written first and verified after the fix. |
| allowed-tools | ["Bash","Read","Edit","Write","Grep","Glob","Agent","AskUserQuestion"] |
Debug the user's bug using a Test-Driven Development approach combined with the collab-fix protocol.
code-reviewer subagent if available; otherwise use the general-purpose agent.test-automator subagent for test design review if available; otherwise use the general-purpose agent.timeout: 1200000 (20 min) when calling Bash for codex/gemini commands.Expand the bug description with context: expected behavior, actual (buggy) behavior, relevant code paths, the project's testing framework. Keep the user's wording verbatim alongside your notes.
Ask codex, gemini-cli, and a test-automator subagent to each propose a reproducing test:
echo "Given this bug: <description>. Design a test case that will FAIL when the bug exists and PASS when fixed. Include the test code and explain why it catches this bug." | codex exec --skip-git-repo-check --sandbox read-only - 2>/dev/nullgemini "Given this bug: <description>. Design a test case that will FAIL when the bug exists and PASS when fixed. Include the test code and explain why it catches this bug." -o json 2>/dev/null | jq -r '.response'AskUserQuestion to let the user pick.AskUserQuestion:
echo "Analyze this bug: <description>. We have a failing test that reproduces it. Propose a fix plan with steps and tradeoffs. The fix must make the test pass." | codex exec --skip-git-repo-check --sandbox read-only - 2>/dev/nullgemini "Analyze this bug: <description>. We have a failing test that reproduces it. Propose a fix plan with steps and tradeoffs. The fix must make the test pass." -o json 2>/dev/null | jq -r '.response'code-reviewer agent to propose a fix independentlyCompare the 3 plans, summarize tradeoffs, and ask the user only the necessary questions via AskUserQuestion.
Ultrathink and implement the fix yourself. Do not git commit.
Run the reproducing test. If it still fails, analyze and iterate.
(echo "Review the following uncommitted diff. Verify: 1) The fix is correct, 2) The test is feasible and properly validates the fix, 3) No regressions introduced."; git diff) | codex exec --skip-git-repo-check --sandbox read-only - 2>/dev/null(echo "Review the following uncommitted diff. Verify: 1) The fix is correct, 2) The test is feasible and properly validates the fix, 3) No regressions introduced."; git diff) | gemini -o json 2>/dev/null | jq -r '.response'code-reviewer agent focused on test validityResolve human-preference items via AskUserQuestion. Repeat 2.3–2.5 until all three reviewers are satisfied AND the test passes, or 5 rounds are reached.
If no consensus after 5 rounds, report:
After completion, report: